1 ================================
2 Source Level Debugging with LLVM
3 ================================
5 .. sectionauthor:: Chris Lattner <sabre@nondot.org> and Jim Laskey <jlaskey@mac.com>
13 This document is the central repository for all information pertaining to debug
14 information in LLVM. It describes the :ref:`actual format that the LLVM debug
15 information takes <format>`, which is useful for those interested in creating
16 front-ends or dealing directly with the information. Further, this document
17 provides specific examples of what debug information for C/C++ looks like.
19 Philosophy behind LLVM debugging information
20 --------------------------------------------
22 The idea of the LLVM debugging information is to capture how the important
23 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
24 Several design aspects have shaped the solution that appears here. The
27 * Debugging information should have very little impact on the rest of the
28 compiler. No transformations, analyses, or code generators should need to
29 be modified because of debugging information.
31 * LLVM optimizations should interact in :ref:`well-defined and easily described
32 ways <intro_debugopt>` with the debugging information.
34 * Because LLVM is designed to support arbitrary programming languages,
35 LLVM-to-LLVM tools should not need to know anything about the semantics of
36 the source-level-language.
38 * Source-level languages are often **widely** different from one another.
39 LLVM should not put any restrictions of the flavor of the source-language,
40 and the debugging information should work with any language.
42 * With code generator support, it should be possible to use an LLVM compiler
43 to compile a program to native machine code and standard debugging
44 formats. This allows compatibility with traditional machine-code level
45 debuggers, like GDB or DBX.
47 The approach used by the LLVM implementation is to use a small set of
48 :ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
49 between LLVM program objects and the source-level objects. The description of
50 the source-level program is maintained in LLVM metadata in an
51 :ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
52 currently uses working draft 7 of the `DWARF 3 standard
53 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
55 When a program is being debugged, a debugger interacts with the user and turns
56 the stored debug information into source-language specific information. As
57 such, a debugger must be aware of the source-language, and is thus tied to a
58 specific language or family of languages.
60 Debug information consumers
61 ---------------------------
63 The role of debug information is to provide meta information normally stripped
64 away during the compilation process. This meta information provides an LLVM
65 user a relationship between generated code and the original program source
68 Currently, debug information is consumed by DwarfDebug to produce dwarf
69 information used by the gdb debugger. Other targets could use the same
70 information to produce stabs or other debug forms.
72 It would also be reasonable to use debug information to feed profiling tools
73 for analysis of generated code, or, tools for reconstructing the original
74 source from generated code.
76 TODO - expound a bit more.
80 Debugging optimized code
81 ------------------------
83 An extremely high priority of LLVM debugging information is to make it interact
84 well with optimizations and analysis. In particular, the LLVM debug
85 information provides the following guarantees:
87 * LLVM debug information **always provides information to accurately read
88 the source-level state of the program**, regardless of which LLVM
89 optimizations have been run, and without any modification to the
90 optimizations themselves. However, some optimizations may impact the
91 ability to modify the current state of the program with a debugger, such
92 as setting program variables, or calling functions that have been
95 * As desired, LLVM optimizations can be upgraded to be aware of the LLVM
96 debugging information, allowing them to update the debugging information
97 as they perform aggressive optimizations. This means that, with effort,
98 the LLVM optimizers could optimize debug code just as well as non-debug
101 * LLVM debug information does not prevent optimizations from
102 happening (for example inlining, basic block reordering/merging/cleanup,
103 tail duplication, etc).
105 * LLVM debug information is automatically optimized along with the rest of
106 the program, using existing facilities. For example, duplicate
107 information is automatically merged by the linker, and unused information
108 is automatically removed.
110 Basically, the debug information allows you to compile a program with
111 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
112 the program as it executes from a debugger. Compiling a program with
113 "``-O3 -g``" gives you full debug information that is always available and
114 accurate for reading (e.g., you get accurate stack traces despite tail call
115 elimination and inlining), but you might lose the ability to modify the program
116 and call functions where were optimized out of the program, or inlined away
119 :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
120 optimizer's handling of debugging information. It can be run like this:
124 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
127 This will test impact of debugging information on optimization passes. If
128 debugging information influences optimization passes then it will be reported
129 as a failure. See :doc:`TestingGuide` for more information on LLVM test
130 infrastructure and how to run various tests.
134 Debugging information format
135 ============================
137 LLVM debugging information has been carefully designed to make it possible for
138 the optimizer to optimize the program and debugging information without
139 necessarily having to know anything about debugging information. In
140 particular, the use of metadata avoids duplicated debugging information from
141 the beginning, and the global dead code elimination pass automatically deletes
142 debugging information for a function if it decides to delete the function.
144 To do this, most of the debugging information (descriptors for types,
145 variables, functions, source files, etc) is inserted by the language front-end
146 in the form of LLVM metadata.
148 Debug information is designed to be agnostic about the target debugger and
149 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
150 pass to decode the information that represents variables, types, functions,
151 namespaces, etc: this allows for arbitrary source-language semantics and
152 type-systems to be used, as long as there is a module written for the target
153 debugger to interpret the information.
155 To provide basic functionality, the LLVM debugger does have to make some
156 assumptions about the source-level language being debugged, though it keeps
157 these to a minimum. The only common features that the LLVM debugger assumes
158 exist are :ref:`source files <format_files>`, and :ref:`program objects
159 <format_global_variables>`. These abstract objects are used by a debugger to
160 form stack traces, show information about local variables, etc.
162 This section of the documentation first describes the representation aspects
163 common to any source-language. :ref:`ccxx_frontend` describes the data layout
164 conventions used by the C and C++ front-ends.
166 Debug information descriptors
167 -----------------------------
169 In consideration of the complexity and volume of debug information, LLVM
170 provides a specification for well formed debug descriptors.
172 Consumers of LLVM debug information expect the descriptors for program objects
173 to start in a canonical format, but the descriptors can include additional
174 information appended at the end that is source-language specific. All LLVM
175 debugging information is versioned, allowing backwards compatibility in the
176 case that the core structures need to change in some way. Also, all debugging
177 information objects start with a tag to indicate what type of object it is.
178 The source-language is allowed to define its own objects, by using unreserved
179 tag numbers. We recommend using with tags in the range 0x1000 through 0x2000
180 (there is a defined ``enum DW_TAG_user_base = 0x1000``.)
182 The fields of debug descriptors used internally by LLVM are restricted to only
183 the simple data types ``i32``, ``i1``, ``float``, ``double``, ``mdstring`` and
193 <a name="LLVMDebugVersion">The first field of a descriptor is always an
194 ``i32`` containing a tag value identifying the content of the descriptor.
195 The remaining fields are specific to the descriptor. The values of tags are
196 loosely bound to the tag values of DWARF information entries. However, that
197 does not restrict the use of the information supplied to DWARF targets. To
198 facilitate versioning of debug information, the tag is augmented with the
199 current debug version (``LLVMDebugVersion = 8 << 16`` or 0x80000 or
202 The details of the various descriptors follow.
204 Compile unit descriptors
205 ^^^^^^^^^^^^^^^^^^^^^^^^
210 i32, ;; Tag = 17 + LLVMDebugVersion (DW_TAG_compile_unit)
211 i32, ;; Unused field.
212 i32, ;; DWARF language identifier (ex. DW_LANG_C89)
213 metadata, ;; Source file name
214 metadata, ;; Source file directory (includes trailing slash)
215 metadata ;; Producer (ex. "4.0.1 LLVM (LLVM research group)")
216 i1, ;; True if this is a main compile unit.
217 i1, ;; True if this is optimized.
219 i32 ;; Runtime version
220 metadata ;; List of enums types
221 metadata ;; List of retained types
222 metadata ;; List of subprograms
223 metadata ;; List of global variables
226 These descriptors contain a source language ID for the file (we use the DWARF
227 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
228 ``DW_LANG_Cobol74``, etc), three strings describing the filename, working
229 directory of the compiler, and an identifier string for the compiler that
232 Compile unit descriptors provide the root context for objects declared in a
233 specific compilation unit. File descriptors are defined using this context.
234 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
235 keep track of subprograms, global variables and type information.
245 i32, ;; Tag = 41 + LLVMDebugVersion (DW_TAG_file_type)
246 metadata, ;; Source file name
247 metadata, ;; Source file directory (includes trailing slash)
251 These descriptors contain information for a file. Global variables and top
252 level functions would be defined using this context. File descriptors also
253 provide context for source line correspondence.
255 Each input file is encoded as a separate file descriptor in LLVM debugging
258 .. _format_global_variables:
260 Global variable descriptors
261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
266 i32, ;; Tag = 52 + LLVMDebugVersion (DW_TAG_variable)
267 i32, ;; Unused field.
268 metadata, ;; Reference to context descriptor
270 metadata, ;; Display name (fully qualified C++ name)
271 metadata, ;; MIPS linkage name (for C++)
272 metadata, ;; Reference to file where defined
273 i32, ;; Line number where defined
274 metadata, ;; Reference to type descriptor
275 i1, ;; True if the global is local to compile unit (static)
276 i1, ;; True if the global is defined in the compile unit (not extern)
277 {}* ;; Reference to the global variable
280 These descriptors provide debug information about globals variables. They
281 provide details such as name, type and where the variable is defined. All
282 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
284 .. _format_subprograms:
286 Subprogram descriptors
287 ^^^^^^^^^^^^^^^^^^^^^^
292 i32, ;; Tag = 46 + LLVMDebugVersion (DW_TAG_subprogram)
293 i32, ;; Unused field.
294 metadata, ;; Reference to context descriptor
296 metadata, ;; Display name (fully qualified C++ name)
297 metadata, ;; MIPS linkage name (for C++)
298 metadata, ;; Reference to file where defined
299 i32, ;; Line number where defined
300 metadata, ;; Reference to type descriptor
301 i1, ;; True if the global is local to compile unit (static)
302 i1, ;; True if the global is defined in the compile unit (not extern)
303 i32, ;; Line number where the scope of the subprogram begins
304 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
305 i32, ;; Index into a virtual function
306 metadata, ;; indicates which base type contains the vtable pointer for the
308 i32, ;; Flags - Artifical, Private, Protected, Explicit, Prototyped.
310 Function * , ;; Pointer to LLVM function
311 metadata, ;; Lists function template parameters
312 metadata, ;; Function declaration descriptor
313 metadata ;; List of function variables
316 These descriptors provide debug information about functions, methods and
317 subprograms. They provide details such as name, return types and the source
318 location where the subprogram is defined.
326 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
327 metadata,;; Reference to context descriptor
329 i32, ;; Column number
330 metadata,;; Reference to source file
331 i32 ;; Unique ID to identify blocks from a template function
334 This descriptor provides debug information about nested blocks within a
335 subprogram. The line number and column numbers are used to dinstinguish two
336 lexical blocks at same depth.
341 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
342 metadata ;; Reference to the scope we're annotating with a file change
343 metadata,;; Reference to the file the scope is enclosed in.
346 This descriptor provides a wrapper around a lexical scope to handle file
347 changes in the middle of a lexical block.
349 .. _format_basic_type:
351 Basic type descriptors
352 ^^^^^^^^^^^^^^^^^^^^^^
357 i32, ;; Tag = 36 + LLVMDebugVersion (DW_TAG_base_type)
358 metadata, ;; Reference to context
359 metadata, ;; Name (may be "" for anonymous types)
360 metadata, ;; Reference to file where defined (may be NULL)
361 i32, ;; Line number where defined (may be 0)
363 i64, ;; Alignment in bits
364 i64, ;; Offset in bits
366 i32 ;; DWARF type encoding
369 These descriptors define primitive types used in the code. Example ``int``,
370 ``bool`` and ``float``. The context provides the scope of the type, which is
371 usually the top level. Since basic types are not usually user defined the
372 context and line number can be left as NULL and 0. The size, alignment and
373 offset are expressed in bits and can be 64 bit values. The alignment is used
374 to round the offset when embedded in a :ref:`composite type
375 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
376 The offset is the bit offset if embedded in a :ref:`composite type
377 <format_composite_type>`.
379 The type encoding provides the details of the type. The values are typically
380 one of the following:
388 DW_ATE_signed_char = 6
390 DW_ATE_unsigned_char = 8
392 .. _format_derived_type:
394 Derived type descriptors
395 ^^^^^^^^^^^^^^^^^^^^^^^^
400 i32, ;; Tag (see below)
401 metadata, ;; Reference to context
402 metadata, ;; Name (may be "" for anonymous types)
403 metadata, ;; Reference to file where defined (may be NULL)
404 i32, ;; Line number where defined (may be 0)
406 i64, ;; Alignment in bits
407 i64, ;; Offset in bits
408 i32, ;; Flags to encode attributes, e.g. private
409 metadata, ;; Reference to type derived from
410 metadata, ;; (optional) Name of the Objective C property associated with
411 ;; Objective-C an ivar
412 metadata, ;; (optional) Name of the Objective C property getter selector.
413 metadata, ;; (optional) Name of the Objective C property setter selector.
414 i32 ;; (optional) Objective C property attributes.
417 These descriptors are used to define types derived from other types. The value
418 of the tag varies depending on the meaning. The following are possible tag
423 DW_TAG_formal_parameter = 5
425 DW_TAG_pointer_type = 15
426 DW_TAG_reference_type = 16
428 DW_TAG_const_type = 38
429 DW_TAG_volatile_type = 53
430 DW_TAG_restrict_type = 55
432 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
433 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
434 of the member is the :ref:`derived type <format_derived_type>`.
435 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
436 argument of a subprogram.
438 ``DW_TAG_typedef`` is used to provide a name for the derived type.
440 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
441 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
442 :ref:`derived type <format_derived_type>`.
444 :ref:`Derived type <format_derived_type>` location can be determined from the
445 context and line number. The size, alignment and offset are expressed in bits
446 and can be 64 bit values. The alignment is used to round the offset when
447 embedded in a :ref:`composite type <format_composite_type>` (example to keep
448 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
449 in a :ref:`composite type <format_composite_type>`.
451 Note that the ``void *`` type is expressed as a type derived from NULL.
453 .. _format_composite_type:
455 Composite type descriptors
456 ^^^^^^^^^^^^^^^^^^^^^^^^^^
461 i32, ;; Tag (see below)
462 metadata, ;; Reference to context
463 metadata, ;; Name (may be "" for anonymous types)
464 metadata, ;; Reference to file where defined (may be NULL)
465 i32, ;; Line number where defined (may be 0)
467 i64, ;; Alignment in bits
468 i64, ;; Offset in bits
470 metadata, ;; Reference to type derived from
471 metadata, ;; Reference to array of member descriptors
472 i32 ;; Runtime languages
475 These descriptors are used to define types that are composed of 0 or more
476 elements. The value of the tag varies depending on the meaning. The following
477 are possible tag values:
481 DW_TAG_array_type = 1
482 DW_TAG_enumeration_type = 4
483 DW_TAG_structure_type = 19
484 DW_TAG_union_type = 23
485 DW_TAG_vector_type = 259
486 DW_TAG_subroutine_type = 21
487 DW_TAG_inheritance = 28
489 The vector flag indicates that an array type is a native packed vector.
491 The members of array types (tag = ``DW_TAG_array_type``) or vector types (tag =
492 ``DW_TAG_vector_type``) are :ref:`subrange descriptors <format_subrange>`, each
493 representing the range of subscripts at that level of indexing.
495 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
496 :ref:`enumerator descriptors <format_enumerator>`, each representing the
497 definition of enumeration value for the set. All enumeration type descriptors
498 are collected inside the named metadata ``!llvm.dbg.cu``.
500 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
501 ``DW_TAG_union_type``) types are any one of the :ref:`basic
502 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
503 <format_composite_type>` type descriptors, each representing a field member of
504 the structure or union.
506 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
507 information about base classes, static members and member functions. If a
508 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
509 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
510 of is a :ref:`global variable descriptor <format_global_variables>` then it
511 represents a static member. And, if the member is a :ref:`subprogram
512 descriptor <format_subprograms>` then it represents a member function. For
513 static members and member functions, ``getName()`` returns the members link or
514 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
516 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
517 is the return type for the subroutine. The remaining elements are the formal
518 arguments to the subroutine.
520 :ref:`Composite type <format_composite_type>` location can be determined from
521 the context and line number. The size, alignment and offset are expressed in
522 bits and can be 64 bit values. The alignment is used to round the offset when
523 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
524 keep float doubles on 64 bit boundaries). The offset is the bit offset if
525 embedded in a :ref:`composite type <format_composite_type>`.
535 i32, ;; Tag = 33 + LLVMDebugVersion (DW_TAG_subrange_type)
540 These descriptors are used to define ranges of array subscripts for an array
541 :ref:`composite type <format_composite_type>`. The low value defines the lower
542 bounds typically zero for C/C++. The high value is the upper bounds. Values
543 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
544 the array bounds are not included in generated debugging information.
546 .. _format_enumerator:
548 Enumerator descriptors
549 ^^^^^^^^^^^^^^^^^^^^^^
554 i32, ;; Tag = 40 + LLVMDebugVersion (DW_TAG_enumerator)
559 These descriptors are used to define members of an enumeration :ref:`composite
560 type <format_composite_type>`, it associates the name to the value.
568 i32, ;; Tag (see below)
571 metadata, ;; Reference to file where defined
572 i32, ;; 24 bit - Line number where defined
573 ;; 8 bit - Argument number. 1 indicates 1st argument.
574 metadata, ;; Type descriptor
576 metadata ;; (optional) Reference to inline location
579 These descriptors are used to define variables local to a sub program. The
580 value of the tag depends on the usage of the variable:
584 DW_TAG_auto_variable = 256
585 DW_TAG_arg_variable = 257
586 DW_TAG_return_variable = 258
588 An auto variable is any variable declared in the body of the function. An
589 argument variable is any variable that appears as a formal argument to the
590 function. A return variable is used to track the result of a function and has
591 no source correspondent.
593 The context is either the subprogram or block where the variable is defined.
594 Name the source variable name. Context and line indicate where the variable
595 was defined. Type descriptor defines the declared type of the variable.
597 .. _format_common_intrinsics:
599 Debugger intrinsic functions
600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
602 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
603 provide debug information at various points in generated code.
610 void %llvm.dbg.declare(metadata, metadata)
612 This intrinsic provides information about a local element (e.g., variable).
613 The first argument is metadata holding the alloca for the variable. The second
614 argument is metadata containing a description of the variable.
621 void %llvm.dbg.value(metadata, i64, metadata)
623 This intrinsic provides information when a user source variable is set to a new
624 value. The first argument is the new value (wrapped as metadata). The second
625 argument is the offset in the user source variable where the new value is
626 written. The third argument is metadata containing a description of the user
629 Object lifetimes and scoping
630 ============================
632 In many languages, the local variables in functions can have their lifetimes or
633 scopes limited to a subset of a function. In the C family of languages, for
634 example, variables are only live (readable and writable) within the source
635 block that they are defined in. In functional languages, values are only
636 readable after they have been defined. Though this is a very obvious concept,
637 it is non-trivial to model in LLVM, because it has no notion of scoping in this
638 sense, and does not want to be tied to a language's scoping rules.
640 In order to handle this, the LLVM debug format uses the metadata attached to
641 llvm instructions to encode line number and scoping information. Consider the
642 following C fragment, for example:
656 Compiled to LLVM, this function would be represented like this:
660 define void @foo() nounwind ssp {
662 %X = alloca i32, align 4 ; <i32*> [#uses=4]
663 %Y = alloca i32, align 4 ; <i32*> [#uses=4]
664 %Z = alloca i32, align 4 ; <i32*> [#uses=3]
665 %0 = bitcast i32* %X to {}* ; <{}*> [#uses=1]
666 call void @llvm.dbg.declare(metadata !{i32 * %X}, metadata !0), !dbg !7
667 store i32 21, i32* %X, !dbg !8
668 %1 = bitcast i32* %Y to {}* ; <{}*> [#uses=1]
669 call void @llvm.dbg.declare(metadata !{i32 * %Y}, metadata !9), !dbg !10
670 store i32 22, i32* %Y, !dbg !11
671 %2 = bitcast i32* %Z to {}* ; <{}*> [#uses=1]
672 call void @llvm.dbg.declare(metadata !{i32 * %Z}, metadata !12), !dbg !14
673 store i32 23, i32* %Z, !dbg !15
674 %tmp = load i32* %X, !dbg !16 ; <i32> [#uses=1]
675 %tmp1 = load i32* %Y, !dbg !16 ; <i32> [#uses=1]
676 %add = add nsw i32 %tmp, %tmp1, !dbg !16 ; <i32> [#uses=1]
677 store i32 %add, i32* %Z, !dbg !16
678 %tmp2 = load i32* %Y, !dbg !17 ; <i32> [#uses=1]
679 store i32 %tmp2, i32* %X, !dbg !17
683 declare void @llvm.dbg.declare(metadata, metadata) nounwind readnone
685 !0 = metadata !{i32 459008, metadata !1, metadata !"X",
686 metadata !3, i32 2, metadata !6}; [ DW_TAG_auto_variable ]
687 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
688 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo", metadata !"foo",
689 metadata !"foo", metadata !3, i32 1, metadata !4,
690 i1 false, i1 true}; [DW_TAG_subprogram ]
691 !3 = metadata !{i32 458769, i32 0, i32 12, metadata !"foo.c",
692 metadata !"/private/tmp", metadata !"clang 1.1", i1 true,
693 i1 false, metadata !"", i32 0}; [DW_TAG_compile_unit ]
694 !4 = metadata !{i32 458773, metadata !3, metadata !"", null, i32 0, i64 0, i64 0,
695 i64 0, i32 0, null, metadata !5, i32 0}; [DW_TAG_subroutine_type ]
696 !5 = metadata !{null}
697 !6 = metadata !{i32 458788, metadata !3, metadata !"int", metadata !3, i32 0,
698 i64 32, i64 32, i64 0, i32 0, i32 5}; [DW_TAG_base_type ]
699 !7 = metadata !{i32 2, i32 7, metadata !1, null}
700 !8 = metadata !{i32 2, i32 3, metadata !1, null}
701 !9 = metadata !{i32 459008, metadata !1, metadata !"Y", metadata !3, i32 3,
702 metadata !6}; [ DW_TAG_auto_variable ]
703 !10 = metadata !{i32 3, i32 7, metadata !1, null}
704 !11 = metadata !{i32 3, i32 3, metadata !1, null}
705 !12 = metadata !{i32 459008, metadata !13, metadata !"Z", metadata !3, i32 5,
706 metadata !6}; [ DW_TAG_auto_variable ]
707 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
708 !14 = metadata !{i32 5, i32 9, metadata !13, null}
709 !15 = metadata !{i32 5, i32 5, metadata !13, null}
710 !16 = metadata !{i32 6, i32 5, metadata !13, null}
711 !17 = metadata !{i32 8, i32 3, metadata !1, null}
712 !18 = metadata !{i32 9, i32 1, metadata !2, null}
714 This example illustrates a few important details about LLVM debugging
715 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
716 location information, which are attached to an instruction, are applied
717 together to allow a debugger to analyze the relationship between statements,
718 variable definitions, and the code used to implement the function.
722 call void @llvm.dbg.declare(metadata, metadata !0), !dbg !7
724 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
725 variable ``X``. The metadata ``!dbg !7`` attached to the intrinsic provides
726 scope information for the variable ``X``.
730 !7 = metadata !{i32 2, i32 7, metadata !1, null}
731 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
732 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo",
733 metadata !"foo", metadata !"foo", metadata !3, i32 1,
734 metadata !4, i1 false, i1 true}; [DW_TAG_subprogram ]
736 Here ``!7`` is metadata providing location information. It has four fields:
737 line number, column number, scope, and original scope. The original scope
738 represents inline location if this instruction is inlined inside a caller, and
739 is null otherwise. In this example, scope is encoded by ``!1``. ``!1``
740 represents a lexical block inside the scope ``!2``, where ``!2`` is a
741 :ref:`subprogram descriptor <format_subprograms>`. This way the location
742 information attached to the intrinsics indicates that the variable ``X`` is
743 declared at line number 2 at a function level scope in function ``foo``.
745 Now lets take another example.
749 call void @llvm.dbg.declare(metadata, metadata !12), !dbg !14
751 The second intrinsic ``%llvm.dbg.declare`` encodes debugging information for
752 variable ``Z``. The metadata ``!dbg !14`` attached to the intrinsic provides
753 scope information for the variable ``Z``.
757 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
758 !14 = metadata !{i32 5, i32 9, metadata !13, null}
760 Here ``!14`` indicates that ``Z`` is declared at line number 5 and
761 column number 9 inside of lexical scope ``!13``. The lexical scope itself
762 resides inside of lexical scope ``!1`` described above.
764 The scope information attached with each instruction provides a straightforward
765 way to find instructions covered by a scope.
769 C/C++ front-end specific debug information
770 ==========================================
772 The C and C++ front-ends represent information about the program in a format
773 that is effectively identical to `DWARF 3.0
774 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
775 content. This allows code generators to trivially support native debuggers by
776 generating standard dwarf information, and contains enough information for
777 non-dwarf targets to translate it as needed.
779 This section describes the forms used to represent C and C++ programs. Other
780 languages could pattern themselves after this (which itself is tuned to
781 representing programs in the same way that DWARF 3 does), or they could choose
782 to provide completely different forms if they don't fit into the DWARF model.
783 As support for debugging information gets added to the various LLVM
784 source-language front-ends, the information used should be documented here.
786 The following sections provide examples of various C/C++ constructs and the
787 debug information that would best describe those constructs.
789 C/C++ source file information
790 -----------------------------
792 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
793 directory ``/Users/mine/sources``, the following code:
797 #include "MyHeader.h"
799 int main(int argc, char *argv[]) {
803 a C/C++ front-end would generate the following descriptors:
809 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
814 i32 4, ;; Language Id
815 metadata !"MySource.cpp",
816 metadata !"/Users/mine/sources",
817 metadata !"4.2.1 (Based on Apple Inc. build 5649) (LLVM build 00)",
818 i1 true, ;; Main Compile Unit
819 i1 false, ;; Optimized compile unit
820 metadata !"", ;; Compiler flags
821 i32 0} ;; Runtime version
824 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
828 metadata !"MySource.cpp",
829 metadata !"/Users/mine/sources",
830 metadata !2 ;; Compile unit
834 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
838 metadata !"Myheader.h"
839 metadata !"/Users/mine/sources",
840 metadata !2 ;; Compile unit
845 ``llvm::Instruction`` provides easy access to metadata attached with an
846 instruction. One can extract line number information encoded in LLVM IR using
847 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
851 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
852 DILocation Loc(N); // DILocation is in DebugInfo.h
853 unsigned Line = Loc.getLineNumber();
854 StringRef File = Loc.getFilename();
855 StringRef Dir = Loc.getDirectory();
858 C/C++ global variable information
859 ---------------------------------
861 Given an integer global variable declared as follows:
867 a C/C++ front-end would generate the following descriptors:
872 ;; Define the global itself.
874 %MyGlobal = global int 100
877 ;; List of debug info of globals
881 ;; Define the compile unit.
886 metadata !"foo.cpp", ;; File
887 metadata !"/Volumes/Data/tmp", ;; Directory
888 metadata !"clang version 3.1 ", ;; Producer
889 i1 true, ;; Deprecated field
890 i1 false, ;; "isOptimized"?
891 metadata !"", ;; Flags
892 i32 0, ;; Runtime Version
893 metadata !1, ;; Enum Types
894 metadata !1, ;; Retained Types
895 metadata !1, ;; Subprograms
896 metadata !3 ;; Global Variables
897 } ; [ DW_TAG_compile_unit ]
899 ;; The Array of Global Variables
909 ;; Define the global variable itself.
915 metadata !"MyGlobal", ;; Name
916 metadata !"MyGlobal", ;; Display Name
917 metadata !"", ;; Linkage Name
921 i32 0, ;; IsLocalToUnit
922 i32 1, ;; IsDefinition
923 i32* @MyGlobal ;; LLVM-IR Value
924 } ; [ DW_TAG_variable ]
931 metadata !"foo.cpp", ;; File
932 metadata !"/Volumes/Data/tmp", ;; Directory
934 } ; [ DW_TAG_file_type ]
942 metadata !"int", ;; Name
945 i64 32, ;; Size in Bits
946 i64 32, ;; Align in Bits
950 } ; [ DW_TAG_base_type ]
952 C/C++ function information
953 --------------------------
955 Given a function declared as follows:
959 int main(int argc, char *argv[]) {
963 a C/C++ front-end would generate the following descriptors:
968 ;; Define the anchor for subprograms. Note that the second field of the
969 ;; anchor is 46, which is the same as the tag for subprograms
970 ;; (46 = DW_TAG_subprogram.)
975 metadata !1, ;; Context
976 metadata !"main", ;; Name
977 metadata !"main", ;; Display name
978 metadata !"main", ;; Linkage name
980 i32 1, ;; Line number
982 i1 false, ;; Is local
983 i1 true, ;; Is definition
984 i32 0, ;; Virtuality attribute, e.g. pure virtual function
985 i32 0, ;; Index into virtual table for C++ methods
986 i32 0, ;; Type that holds virtual table.
988 i1 false, ;; True if this function is optimized
989 Function *, ;; Pointer to llvm::Function
990 null ;; Function template parameters
993 ;; Define the subprogram itself.
995 define i32 @main(i32 %argc, i8** %argv) {
1002 The following are the basic type descriptors for C/C++ core types:
1007 .. code-block:: llvm
1011 metadata !1, ;; Context
1012 metadata !"bool", ;; Name
1013 metadata !1, ;; File
1014 i32 0, ;; Line number
1015 i64 8, ;; Size in Bits
1016 i64 8, ;; Align in Bits
1017 i64 0, ;; Offset in Bits
1025 .. code-block:: llvm
1029 metadata !1, ;; Context
1030 metadata !"char", ;; Name
1031 metadata !1, ;; File
1032 i32 0, ;; Line number
1033 i64 8, ;; Size in Bits
1034 i64 8, ;; Align in Bits
1035 i64 0, ;; Offset in Bits
1043 .. code-block:: llvm
1047 metadata !1, ;; Context
1048 metadata !"unsigned char",
1049 metadata !1, ;; File
1050 i32 0, ;; Line number
1051 i64 8, ;; Size in Bits
1052 i64 8, ;; Align in Bits
1053 i64 0, ;; Offset in Bits
1061 .. code-block:: llvm
1065 metadata !1, ;; Context
1066 metadata !"short int",
1067 metadata !1, ;; File
1068 i32 0, ;; Line number
1069 i64 16, ;; Size in Bits
1070 i64 16, ;; Align in Bits
1071 i64 0, ;; Offset in Bits
1079 .. code-block:: llvm
1083 metadata !1, ;; Context
1084 metadata !"short unsigned int",
1085 metadata !1, ;; File
1086 i32 0, ;; Line number
1087 i64 16, ;; Size in Bits
1088 i64 16, ;; Align in Bits
1089 i64 0, ;; Offset in Bits
1097 .. code-block:: llvm
1101 metadata !1, ;; Context
1102 metadata !"int", ;; Name
1103 metadata !1, ;; File
1104 i32 0, ;; Line number
1105 i64 32, ;; Size in Bits
1106 i64 32, ;; Align in Bits
1107 i64 0, ;; Offset in Bits
1115 .. code-block:: llvm
1119 metadata !1, ;; Context
1120 metadata !"unsigned int",
1121 metadata !1, ;; File
1122 i32 0, ;; Line number
1123 i64 32, ;; Size in Bits
1124 i64 32, ;; Align in Bits
1125 i64 0, ;; Offset in Bits
1133 .. code-block:: llvm
1137 metadata !1, ;; Context
1138 metadata !"long long int",
1139 metadata !1, ;; File
1140 i32 0, ;; Line number
1141 i64 64, ;; Size in Bits
1142 i64 64, ;; Align in Bits
1143 i64 0, ;; Offset in Bits
1151 .. code-block:: llvm
1155 metadata !1, ;; Context
1156 metadata !"long long unsigned int",
1157 metadata !1, ;; File
1158 i32 0, ;; Line number
1159 i64 64, ;; Size in Bits
1160 i64 64, ;; Align in Bits
1161 i64 0, ;; Offset in Bits
1169 .. code-block:: llvm
1173 metadata !1, ;; Context
1175 metadata !1, ;; File
1176 i32 0, ;; Line number
1177 i64 32, ;; Size in Bits
1178 i64 32, ;; Align in Bits
1179 i64 0, ;; Offset in Bits
1187 .. code-block:: llvm
1191 metadata !1, ;; Context
1192 metadata !"double",;; Name
1193 metadata !1, ;; File
1194 i32 0, ;; Line number
1195 i64 64, ;; Size in Bits
1196 i64 64, ;; Align in Bits
1197 i64 0, ;; Offset in Bits
1205 Given the following as an example of C/C++ derived type:
1209 typedef const int *IntPtr;
1211 a C/C++ front-end would generate the following descriptors:
1213 .. code-block:: llvm
1216 ;; Define the typedef "IntPtr".
1220 metadata !1, ;; Context
1221 metadata !"IntPtr", ;; Name
1222 metadata !3, ;; File
1223 i32 0, ;; Line number
1224 i64 0, ;; Size in bits
1225 i64 0, ;; Align in bits
1226 i64 0, ;; Offset in bits
1228 metadata !4 ;; Derived From type
1231 ;; Define the pointer type.
1235 metadata !1, ;; Context
1236 metadata !"", ;; Name
1237 metadata !1, ;; File
1238 i32 0, ;; Line number
1239 i64 64, ;; Size in bits
1240 i64 64, ;; Align in bits
1241 i64 0, ;; Offset in bits
1243 metadata !5 ;; Derived From type
1246 ;; Define the const type.
1250 metadata !1, ;; Context
1251 metadata !"", ;; Name
1252 metadata !1, ;; File
1253 i32 0, ;; Line number
1254 i64 32, ;; Size in bits
1255 i64 32, ;; Align in bits
1256 i64 0, ;; Offset in bits
1258 metadata !6 ;; Derived From type
1261 ;; Define the int type.
1265 metadata !1, ;; Context
1266 metadata !"int", ;; Name
1267 metadata !1, ;; File
1268 i32 0, ;; Line number
1269 i64 32, ;; Size in bits
1270 i64 32, ;; Align in bits
1271 i64 0, ;; Offset in bits
1276 C/C++ struct/union types
1277 ------------------------
1279 Given the following as an example of C/C++ struct type:
1289 a C/C++ front-end would generate the following descriptors:
1291 .. code-block:: llvm
1294 ;; Define basic type for unsigned int.
1298 metadata !1, ;; Context
1299 metadata !"unsigned int",
1300 metadata !1, ;; File
1301 i32 0, ;; Line number
1302 i64 32, ;; Size in Bits
1303 i64 32, ;; Align in Bits
1304 i64 0, ;; Offset in Bits
1309 ;; Define composite type for struct Color.
1313 metadata !1, ;; Context
1314 metadata !"Color", ;; Name
1315 metadata !1, ;; Compile unit
1316 i32 1, ;; Line number
1317 i64 96, ;; Size in bits
1318 i64 32, ;; Align in bits
1319 i64 0, ;; Offset in bits
1321 null, ;; Derived From
1322 metadata !3, ;; Elements
1323 i32 0 ;; Runtime Language
1327 ;; Define the Red field.
1331 metadata !1, ;; Context
1332 metadata !"Red", ;; Name
1333 metadata !1, ;; File
1334 i32 2, ;; Line number
1335 i64 32, ;; Size in bits
1336 i64 32, ;; Align in bits
1337 i64 0, ;; Offset in bits
1339 metadata !5 ;; Derived From type
1343 ;; Define the Green field.
1347 metadata !1, ;; Context
1348 metadata !"Green", ;; Name
1349 metadata !1, ;; File
1350 i32 3, ;; Line number
1351 i64 32, ;; Size in bits
1352 i64 32, ;; Align in bits
1353 i64 32, ;; Offset in bits
1355 metadata !5 ;; Derived From type
1359 ;; Define the Blue field.
1363 metadata !1, ;; Context
1364 metadata !"Blue", ;; Name
1365 metadata !1, ;; File
1366 i32 4, ;; Line number
1367 i64 32, ;; Size in bits
1368 i64 32, ;; Align in bits
1369 i64 64, ;; Offset in bits
1371 metadata !5 ;; Derived From type
1375 ;; Define the array of fields used by the composite type Color.
1377 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1379 C/C++ enumeration types
1380 -----------------------
1382 Given the following as an example of C/C++ enumeration type:
1392 a C/C++ front-end would generate the following descriptors:
1394 .. code-block:: llvm
1397 ;; Define composite type for enum Trees
1401 metadata !1, ;; Context
1402 metadata !"Trees", ;; Name
1403 metadata !1, ;; File
1404 i32 1, ;; Line number
1405 i64 32, ;; Size in bits
1406 i64 32, ;; Align in bits
1407 i64 0, ;; Offset in bits
1409 null, ;; Derived From type
1410 metadata !3, ;; Elements
1411 i32 0 ;; Runtime language
1415 ;; Define the array of enumerators used by composite type Trees.
1417 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1420 ;; Define Spruce enumerator.
1422 !4 = metadata !{i32 524328, metadata !"Spruce", i64 100}
1425 ;; Define Oak enumerator.
1427 !5 = metadata !{i32 524328, metadata !"Oak", i64 200}
1430 ;; Define Maple enumerator.
1432 !6 = metadata !{i32 524328, metadata !"Maple", i64 300}
1434 Debugging information format
1435 ============================
1437 Debugging Information Extension for Objective C Properties
1438 ----------------------------------------------------------
1443 Objective C provides a simpler way to declare and define accessor methods using
1444 declared properties. The language provides features to declare a property and
1445 to let compiler synthesize accessor methods.
1447 The debugger lets developer inspect Objective C interfaces and their instance
1448 variables and class variables. However, the debugger does not know anything
1449 about the properties defined in Objective C interfaces. The debugger consumes
1450 information generated by compiler in DWARF format. The format does not support
1451 encoding of Objective C properties. This proposal describes DWARF extensions to
1452 encode Objective C properties, which the debugger can use to let developers
1453 inspect Objective C properties.
1458 Objective C properties exist separately from class members. A property can be
1459 defined only by "setter" and "getter" selectors, and be calculated anew on each
1460 access. Or a property can just be a direct access to some declared ivar.
1461 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1462 in which case the property can be referred to in user code directly using the
1463 standard C dereference syntax as well as through the property "dot" syntax, but
1464 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1466 To facilitate debugging, these properties we will add a new DWARF TAG into the
1467 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1468 given property, and a set of DWARF attributes that provide said description.
1469 The property tag will also contain the name and declared type of the property.
1471 If there is a related ivar, there will also be a DWARF property attribute placed
1472 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1473 for that property. And in the case where the compiler synthesizes the ivar
1474 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1475 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1476 to access this ivar directly in code, and with the property attribute pointing
1477 back to the property it is backing.
1479 The following examples will serve as illustration for our discussion:
1481 .. code-block:: objc
1493 @synthesize p2 = n2;
1496 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1498 .. code-block:: none
1500 0x00000100: TAG_structure_type [7] *
1501 AT_APPLE_runtime_class( 0x10 )
1503 AT_decl_file( "Objc_Property.m" )
1506 0x00000110 TAG_APPLE_property
1508 AT_type ( {0x00000150} ( int ) )
1510 0x00000120: TAG_APPLE_property
1512 AT_type ( {0x00000150} ( int ) )
1514 0x00000130: TAG_member [8]
1516 AT_APPLE_property ( {0x00000110} "p1" )
1517 AT_type( {0x00000150} ( int ) )
1518 AT_artificial ( 0x1 )
1520 0x00000140: TAG_member [8]
1522 AT_APPLE_property ( {0x00000120} "p2" )
1523 AT_type( {0x00000150} ( int ) )
1525 0x00000150: AT_type( ( int ) )
1527 Note, the current convention is that the name of the ivar for an
1528 auto-synthesized property is the name of the property from which it derives
1529 with an underscore prepended, as is shown in the example. But we actually
1530 don't need to know this convention, since we are given the name of the ivar
1533 Also, it is common practice in ObjC to have different property declarations in
1534 the @interface and @implementation - e.g. to provide a read-only property in
1535 the interface,and a read-write interface in the implementation. In that case,
1536 the compiler should emit whichever property declaration will be in force in the
1537 current translation unit.
1539 Developers can decorate a property with attributes which are encoded using
1540 ``DW_AT_APPLE_property_attribute``.
1542 .. code-block:: objc
1544 @property (readonly, nonatomic) int pr;
1546 .. code-block:: none
1548 TAG_APPLE_property [8]
1550 AT_type ( {0x00000147} (int) )
1551 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1553 The setter and getter method names are attached to the property using
1554 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1556 .. code-block:: objc
1559 @property (setter=myOwnP3Setter:) int p3;
1560 -(void)myOwnP3Setter:(int)a;
1565 -(void)myOwnP3Setter:(int)a{ }
1568 The DWARF for this would be:
1570 .. code-block:: none
1572 0x000003bd: TAG_structure_type [7] *
1573 AT_APPLE_runtime_class( 0x10 )
1575 AT_decl_file( "Objc_Property.m" )
1578 0x000003cd TAG_APPLE_property
1580 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1581 AT_type( {0x00000147} ( int ) )
1583 0x000003f3: TAG_member [8]
1585 AT_type ( {0x00000147} ( int ) )
1586 AT_APPLE_property ( {0x000003cd} )
1587 AT_artificial ( 0x1 )
1592 +-----------------------+--------+
1594 +=======================+========+
1595 | DW_TAG_APPLE_property | 0x4200 |
1596 +-----------------------+--------+
1598 New DWARF Attributes
1599 ^^^^^^^^^^^^^^^^^^^^
1601 +--------------------------------+--------+-----------+
1602 | Attribute | Value | Classes |
1603 +================================+========+===========+
1604 | DW_AT_APPLE_property | 0x3fed | Reference |
1605 +--------------------------------+--------+-----------+
1606 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1607 +--------------------------------+--------+-----------+
1608 | DW_AT_APPLE_property_setter | 0x3fea | String |
1609 +--------------------------------+--------+-----------+
1610 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1611 +--------------------------------+--------+-----------+
1616 +--------------------------------+-------+
1618 +================================+=======+
1619 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1620 +--------------------------------+-------+
1621 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1622 +--------------------------------+-------+
1623 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1624 +--------------------------------+-------+
1625 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1626 +--------------------------------+-------+
1627 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1628 +--------------------------------+-------+
1629 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1630 +--------------------------------+-------+
1632 Name Accelerator Tables
1633 -----------------------
1638 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1639 debugger needs. The "``pub``" in the section name indicates that the entries
1640 in the table are publicly visible names only. This means no static or hidden
1641 functions show up in the "``.debug_pubnames``". No static variables or private
1642 class variables are in the "``.debug_pubtypes``". Many compilers add different
1643 things to these tables, so we can't rely upon the contents between gcc, icc, or
1646 The typical query given by users tends not to match up with the contents of
1647 these tables. For example, the DWARF spec states that "In the case of the name
1648 of a function member or static data member of a C++ structure, class or union,
1649 the name presented in the "``.debug_pubnames``" section is not the simple name
1650 given by the ``DW_AT_name attribute`` of the referenced debugging information
1651 entry, but rather the fully qualified name of the data or function member."
1652 So the only names in these tables for complex C++ entries is a fully
1653 qualified name. Debugger users tend not to enter their search strings as
1654 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1655 "``a::b::c``". So the name entered in the name table must be demangled in
1656 order to chop it up appropriately and additional names must be manually entered
1657 into the table to make it effective as a name lookup table for debuggers to
1660 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1661 its inconsistent and useless public-only name content making it a waste of
1662 space in the object file. These tables, when they are written to disk, are not
1663 sorted in any way, leaving every debugger to do its own parsing and sorting.
1664 These tables also include an inlined copy of the string values in the table
1665 itself making the tables much larger than they need to be on disk, especially
1666 for large C++ programs.
1668 Can't we just fix the sections by adding all of the names we need to this
1669 table? No, because that is not what the tables are defined to contain and we
1670 won't know the difference between the old bad tables and the new good tables.
1671 At best we could make our own renamed sections that contain all of the data we
1674 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1675 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1676 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1677 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1678 tables. Since clang asks a lot of questions when it is parsing an expression,
1679 we need to be very fast when looking up names, as it happens a lot. Having new
1680 accelerator tables that are optimized for very quick lookups will benefit this
1681 type of debugging experience greatly.
1683 We would like to generate name lookup tables that can be mapped into memory
1684 from disk, and used as is, with little or no up-front parsing. We would also
1685 be able to control the exact content of these different tables so they contain
1686 exactly what we need. The Name Accelerator Tables were designed to fix these
1687 issues. In order to solve these issues we need to:
1689 * Have a format that can be mapped into memory from disk and used as is
1690 * Lookups should be very fast
1691 * Extensible table format so these tables can be made by many producers
1692 * Contain all of the names needed for typical lookups out of the box
1693 * Strict rules for the contents of tables
1695 Table size is important and the accelerator table format should allow the reuse
1696 of strings from common string tables so the strings for the names are not
1697 duplicated. We also want to make sure the table is ready to be used as-is by
1698 simply mapping the table into memory with minimal header parsing.
1700 The name lookups need to be fast and optimized for the kinds of lookups that
1701 debuggers tend to do. Optimally we would like to touch as few parts of the
1702 mapped table as possible when doing a name lookup and be able to quickly find
1703 the name entry we are looking for, or discover there are no matches. In the
1704 case of debuggers we optimized for lookups that fail most of the time.
1706 Each table that is defined should have strict rules on exactly what is in the
1707 accelerator tables and documented so clients can rely on the content.
1712 Standard Hash Tables
1713 """"""""""""""""""""
1715 Typical hash tables have a header, buckets, and each bucket points to the
1718 .. code-block:: none
1728 The BUCKETS are an array of offsets to DATA for each hash:
1730 .. code-block:: none
1733 | 0x00001000 | BUCKETS[0]
1734 | 0x00002000 | BUCKETS[1]
1735 | 0x00002200 | BUCKETS[2]
1736 | 0x000034f0 | BUCKETS[3]
1738 | 0xXXXXXXXX | BUCKETS[n_buckets]
1741 So for ``bucket[3]`` in the example above, we have an offset into the table
1742 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1743 contain a next pointer, full 32 bit hash value, the string itself, and the data
1744 for the current string value.
1746 .. code-block:: none
1749 0x000034f0: | 0x00003500 | next pointer
1750 | 0x12345678 | 32 bit hash
1751 | "erase" | string value
1752 | data[n] | HashData for this bucket
1754 0x00003500: | 0x00003550 | next pointer
1755 | 0x29273623 | 32 bit hash
1756 | "dump" | string value
1757 | data[n] | HashData for this bucket
1759 0x00003550: | 0x00000000 | next pointer
1760 | 0x82638293 | 32 bit hash
1761 | "main" | string value
1762 | data[n] | HashData for this bucket
1765 The problem with this layout for debuggers is that we need to optimize for the
1766 negative lookup case where the symbol we're searching for is not present. So
1767 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1768 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1769 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1770 so, we need to read the next pointer, then read the hash, compare it, and skip
1771 to the next bucket. Each time we are skipping many bytes in memory and
1772 touching new cache pages just to do the compare on the full 32 bit hash. All
1773 of these accesses then tell us that we didn't have a match.
1778 To solve the issues mentioned above we have structured the hash tables a bit
1779 differently: a header, buckets, an array of all unique 32 bit hash values,
1780 followed by an array of hash value data offsets, one for each hash value, then
1781 the data for all hash values:
1783 .. code-block:: none
1797 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1798 making all of the full 32 bit hash values contiguous in memory, we allow
1799 ourselves to efficiently check for a match while touching as little memory as
1800 possible. Most often checking the 32 bit hash values is as far as the lookup
1801 goes. If it does match, it usually is a match with no collisions. So for a
1802 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1803 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1806 .. code-block:: none
1808 .-------------------------.
1809 | HEADER.magic | uint32_t
1810 | HEADER.version | uint16_t
1811 | HEADER.hash_function | uint16_t
1812 | HEADER.bucket_count | uint32_t
1813 | HEADER.hashes_count | uint32_t
1814 | HEADER.header_data_len | uint32_t
1815 | HEADER_DATA | HeaderData
1816 |-------------------------|
1817 | BUCKETS | uint32_t[bucket_count] // 32 bit hash indexes
1818 |-------------------------|
1819 | HASHES | uint32_t[hashes_count] // 32 bit hash values
1820 |-------------------------|
1821 | OFFSETS | uint32_t[hashes_count] // 32 bit offsets to hash value data
1822 |-------------------------|
1824 `-------------------------'
1826 So taking the exact same data from the standard hash example above we end up
1829 .. code-block:: none
1839 | ... | BUCKETS[n_buckets]
1841 | 0x........ | HASHES[0]
1842 | 0x........ | HASHES[1]
1843 | 0x........ | HASHES[2]
1844 | 0x........ | HASHES[3]
1845 | 0x........ | HASHES[4]
1846 | 0x........ | HASHES[5]
1847 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1848 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1849 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1850 | 0x........ | HASHES[9]
1851 | 0x........ | HASHES[10]
1852 | 0x........ | HASHES[11]
1853 | 0x........ | HASHES[12]
1854 | 0x........ | HASHES[13]
1855 | 0x........ | HASHES[n_hashes]
1857 | 0x........ | OFFSETS[0]
1858 | 0x........ | OFFSETS[1]
1859 | 0x........ | OFFSETS[2]
1860 | 0x........ | OFFSETS[3]
1861 | 0x........ | OFFSETS[4]
1862 | 0x........ | OFFSETS[5]
1863 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1864 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1865 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1866 | 0x........ | OFFSETS[9]
1867 | 0x........ | OFFSETS[10]
1868 | 0x........ | OFFSETS[11]
1869 | 0x........ | OFFSETS[12]
1870 | 0x........ | OFFSETS[13]
1871 | 0x........ | OFFSETS[n_hashes]
1879 0x000034f0: | 0x00001203 | .debug_str ("erase")
1880 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1881 | 0x........ | HashData[0]
1882 | 0x........ | HashData[1]
1883 | 0x........ | HashData[2]
1884 | 0x........ | HashData[3]
1885 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1887 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1888 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1889 | 0x........ | HashData[0]
1890 | 0x........ | HashData[1]
1891 | 0x00001203 | String offset into .debug_str ("dump")
1892 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1893 | 0x........ | HashData[0]
1894 | 0x........ | HashData[1]
1895 | 0x........ | HashData[2]
1896 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1898 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1899 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1900 | 0x........ | HashData[0]
1901 | 0x........ | HashData[1]
1902 | 0x........ | HashData[2]
1903 | 0x........ | HashData[3]
1904 | 0x........ | HashData[4]
1905 | 0x........ | HashData[5]
1906 | 0x........ | HashData[6]
1907 | 0x........ | HashData[7]
1908 | 0x........ | HashData[8]
1909 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1912 So we still have all of the same data, we just organize it more efficiently for
1913 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1914 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1915 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1916 is the index into the ``HASHES`` table. We would then compare any consecutive
1917 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1918 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1919 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1920 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1921 before we know that we have no match. We don't end up marching through
1922 multiple words of memory and we really keep the number of processor data cache
1923 lines being accessed as small as possible.
1925 The string hash that is used for these lookup tables is the Daniel J.
1926 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1927 very good hash for all kinds of names in programs with very few hash
1930 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1935 These name hash tables are designed to be generic where specializations of the
1936 table get to define additional data that goes into the header ("``HeaderData``"),
1937 how the string value is stored ("``KeyType``") and the content of the data for each
1943 The header has a fixed part, and the specialized part. The exact format of the
1950 uint32_t magic; // 'HASH' magic value to allow endian detection
1951 uint16_t version; // Version number
1952 uint16_t hash_function; // The hash function enumeration that was used
1953 uint32_t bucket_count; // The number of buckets in this hash table
1954 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1955 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1956 // Specifically the length of the following HeaderData field - this does not
1957 // include the size of the preceding fields
1958 HeaderData header_data; // Implementation specific header data
1961 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1962 encoded as an ASCII integer. This allows the detection of the start of the
1963 hash table and also allows the table's byte order to be determined so the table
1964 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1965 ``version`` number which allows the table to be revised and modified in the
1966 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1967 enumeration that specifies which hash function was used to produce this table.
1968 The current values for the hash function enumerations include:
1972 enum HashFunctionType
1974 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1977 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1978 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1979 hash values that are in the ``HASHES`` array, and is the same number of offsets
1980 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1981 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1987 The header is followed by the buckets, hashes, offsets, and hash value data.
1993 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1994 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1995 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
1998 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
1999 ``hashes`` array contains all of the 32 bit hash values for all names in the
2000 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
2001 array that points to the data for the hash value.
2003 This table setup makes it very easy to repurpose these tables to contain
2004 different data, while keeping the lookup mechanism the same for all tables.
2005 This layout also makes it possible to save the table to disk and map it in
2006 later and do very efficient name lookups with little or no parsing.
2008 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2009 of information for each name. We want to make the DWARF tables extensible and
2010 able to store the data efficiently so we have used some of the DWARF features
2011 that enable efficient data storage to define exactly what kind of data we store
2014 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2015 We might want to store an offset to all of the debug information entries (DIEs)
2016 for each name. To keep things extensible, we create a list of items, or
2017 Atoms, that are contained in the data for each name. First comes the type of
2018 the data in each atom:
2025 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2026 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2027 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2028 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2029 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2032 The enumeration values and their meanings are:
2034 .. code-block:: none
2036 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2037 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2038 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2039 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2040 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2041 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2043 Then we allow each atom type to define the atom type and how the data for each
2044 atom type data is encoded:
2050 uint16_t type; // AtomType enum value
2051 uint16_t form; // DWARF DW_FORM_XXX defines
2054 The ``form`` type above is from the DWARF specification and defines the exact
2055 encoding of the data for the Atom type. See the DWARF specification for the
2056 ``DW_FORM_`` definitions.
2062 uint32_t die_offset_base;
2063 uint32_t atom_count;
2064 Atoms atoms[atom_count0];
2067 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2068 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2069 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2070 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2071 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2072 should be interpreted.
2074 For the current implementations of the "``.apple_names``" (all functions +
2075 globals), the "``.apple_types``" (names of all types that are defined), and
2076 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2081 HeaderData.atom_count = 1;
2082 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2083 HeaderData.atoms[0].form = DW_FORM_data4;
2085 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2086 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2087 multiple matching DIEs in a single file, which could come up with an inlined
2088 function for instance. Future tables could include more information about the
2089 DIE such as flags indicating if the DIE is a function, method, block,
2092 The KeyType for the DWARF table is a 32 bit string table offset into the
2093 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2094 may already contain copies of all of the strings. This helps make sure, with
2095 help from the compiler, that we reuse the strings between all of the DWARF
2096 sections and keeps the hash table size down. Another benefit to having the
2097 compiler generate all strings as DW_FORM_strp in the debug info, is that
2098 DWARF parsing can be made much faster.
2100 After a lookup is made, we get an offset into the hash data. The hash data
2101 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2102 at the offset in the hash data consists of a triple:
2107 uint32_t hash_data_count
2108 HashData[hash_data_count]
2110 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2111 hash data chunks contain a single item (no 32 bit hash collision):
2113 .. code-block:: none
2116 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2117 | 0x00000004 | uint32_t HashData count
2118 | 0x........ | uint32_t HashData[0] DIE offset
2119 | 0x........ | uint32_t HashData[1] DIE offset
2120 | 0x........ | uint32_t HashData[2] DIE offset
2121 | 0x........ | uint32_t HashData[3] DIE offset
2122 | 0x00000000 | uint32_t KeyType (end of hash chain)
2125 If there are collisions, you will have multiple valid string offsets:
2127 .. code-block:: none
2130 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2131 | 0x00000004 | uint32_t HashData count
2132 | 0x........ | uint32_t HashData[0] DIE offset
2133 | 0x........ | uint32_t HashData[1] DIE offset
2134 | 0x........ | uint32_t HashData[2] DIE offset
2135 | 0x........ | uint32_t HashData[3] DIE offset
2136 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2137 | 0x00000002 | uint32_t HashData count
2138 | 0x........ | uint32_t HashData[0] DIE offset
2139 | 0x........ | uint32_t HashData[1] DIE offset
2140 | 0x00000000 | uint32_t KeyType (end of hash chain)
2143 Current testing with real world C++ binaries has shown that there is around 1
2144 32 bit hash collision per 100,000 name entries.
2149 As we said, we want to strictly define exactly what is included in the
2150 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2151 "``.apple_types``", and "``.apple_namespaces``".
2153 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2154 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2155 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2156 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2157 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2158 static variables). All global and static variables should be included,
2159 including those scoped within functions and classes. For example using the
2171 Both of the static ``var`` variables would be included in the table. All
2172 functions should emit both their full names and their basenames. For C or C++,
2173 the full name is the mangled name (if available) which is usually in the
2174 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2175 function basename. If global or static variables have a mangled name in a
2176 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2177 simple name found in the ``DW_AT_name`` attribute.
2179 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2184 * DW_TAG_enumeration_type
2185 * DW_TAG_pointer_type
2186 * DW_TAG_reference_type
2187 * DW_TAG_string_type
2188 * DW_TAG_structure_type
2189 * DW_TAG_subroutine_type
2192 * DW_TAG_ptr_to_member_type
2194 * DW_TAG_subrange_type
2200 * DW_TAG_packed_type
2201 * DW_TAG_volatile_type
2202 * DW_TAG_restrict_type
2203 * DW_TAG_interface_type
2204 * DW_TAG_unspecified_type
2205 * DW_TAG_shared_type
2207 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2208 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2209 value). For example, using the following code:
2219 We get a few type DIEs:
2221 .. code-block:: none
2223 0x00000067: TAG_base_type [5]
2224 AT_encoding( DW_ATE_signed )
2226 AT_byte_size( 0x04 )
2228 0x0000006e: TAG_pointer_type [6]
2229 AT_type( {0x00000067} ( int ) )
2230 AT_byte_size( 0x08 )
2232 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2234 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2235 If we run into a namespace that has no name this is an anonymous namespace, and
2236 the name should be output as "``(anonymous namespace)``" (without the quotes).
2237 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2238 standard C++ library that demangles mangled names.
2241 Language Extensions and File Format Changes
2242 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2244 Objective-C Extensions
2245 """"""""""""""""""""""
2247 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2248 Objective-C class. The name used in the hash table is the name of the
2249 Objective-C class itself. If the Objective-C class has a category, then an
2250 entry is made for both the class name without the category, and for the class
2251 name with the category. So if we have a DIE at offset 0x1234 with a name of
2252 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2253 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2254 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2255 track down all Objective-C methods for an Objective-C class when doing
2256 expressions. It is needed because of the dynamic nature of Objective-C where
2257 anyone can add methods to a class. The DWARF for Objective-C methods is also
2258 emitted differently from C++ classes where the methods are not usually
2259 contained in the class definition, they are scattered about across one or more
2260 compile units. Categories can also be defined in different shared libraries.
2261 So we need to be able to quickly find all of the methods and class functions
2262 given the Objective-C class name, or quickly find all methods and class
2263 functions for a class + category name. This table does not contain any
2264 selector names, it just maps Objective-C class names (or class names +
2265 category) to all of the methods and class functions. The selectors are added
2266 as function basenames in the "``.debug_names``" section.
2268 In the "``.apple_names``" section for Objective-C functions, the full name is
2269 the entire function name with the brackets ("``-[NSString
2270 stringWithCString:]``") and the basename is the selector only
2271 ("``stringWithCString:``").
2276 The sections names for the apple hash tables are for non mach-o files. For
2277 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2280 * "``.apple_names``" -> "``__apple_names``"
2281 * "``.apple_types``" -> "``__apple_types``"
2282 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2283 * "``.apple_objc``" -> "``__apple_objc``"